Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77387.36857470869,
-      "required_pages_per_second_upper_estimate": 0.005353837721432454
+      "required_token_positions_per_second": 77327.21616184319,
+      "required_pages_per_second_upper_estimate": 0.005349676237937591
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9673.421071838586,
-      "required_pages_per_second_upper_estimate": 0.004050059398776195
+      "required_token_positions_per_second": 9665.902020230398,
+      "required_pages_per_second_upper_estimate": 0.0040469113288835
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19346.84214367717,
-      "required_pages_per_second_upper_estimate": 0.004029437007990453
+      "required_token_positions_per_second": 19331.804040460796,
+      "required_pages_per_second_upper_estimate": 0.0040263049676719335
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29020.263215515755,
-      "required_pages_per_second_upper_estimate": 0.0017397204683939514
+      "required_token_positions_per_second": 28997.7060606912,
+      "required_pages_per_second_upper_estimate": 0.0017383682014050002
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29020.263215515755,
-      "required_pages_per_second_upper_estimate": 0.0022561978183585785
+      "required_token_positions_per_second": 28997.7060606912,
+      "required_pages_per_second_upper_estimate": 0.0022544440987894076
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29020.263215515755,
-      "required_pages_per_second_upper_estimate": 0.37224077699767516
+      "required_token_positions_per_second": 28997.7060606912,
+      "required_pages_per_second_upper_estimate": 0.3719514380355716
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1883033916093666,
-    "producer_p95_seconds": 0.2410559790005209,
-    "gpu_update_seconds": 0.952713619262348,
-    "producer_batches_per_second": 5.31057880292719,
-    "gpu_batches_per_second": 1.0496333628297077,
-    "producer_headroom": 5.059460751714671,
+    "producer_batch_seconds": 0.06381905176559144,
+    "producer_p95_seconds": 0.22519886599911842,
+    "gpu_update_seconds": 0.9534547299063482,
+    "producer_batches_per_second": 15.669302071002535,
+    "gpu_batches_per_second": 1.0488174935145833,
+    "producer_headroom": 14.939970173928705,
     "sustainable": true,
-    "buffer_seconds": 60.020958013527924,
+    "buffer_seconds": 60.067647984099935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365587.72700647527,
+    "generation_bytes_per_second": 365337.9726479339,
     "upload_bytes_per_second": 5467407.566198214,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.44661576886838,
+    "estimated_full_shards_per_hour": 78.39302429750931,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 545900, "loss": 0.5648422241210938, "lr": 0.0003, "grad_norm": 0.1479416133957257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79902905568, "tokens_per_second": 153692.59561944264, "data_wait_s": 1.0698000551201403e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 546000, "loss": 0.5812898874282837, "lr": 0.0003, "grad_norm": 0.20790069483774973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79917449390, "tokens_per_second": 151440.93368076577, "data_wait_s": 1.1511998309288174e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 546100, "loss": 0.599733293056488, "lr": 0.0003, "grad_norm": 0.16453615992413548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79931965941, "tokens_per_second": 151729.16107649906, "data_wait_s": 1.5192999853752553e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 546200, "loss": 0.5467432737350464, "lr": 0.0003, "grad_norm": 0.16361261246496153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79946568878, "tokens_per_second": 152129.20605654106, "data_wait_s": 1.5665998944314197e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:140e5a40f409e1e89cc52c1a950d37a68c088f25de21c360050bddec460d9edb
-size 450195
+oid sha256:b5131373d64396c55aa0e375c6f9e59bcda37517828eccc4c0a465c3518b0a46
+size 452313